I don't think it's that hard to run a railroad-free Lebowski adventure--however, what is hard is to publish one. The problem is: a published scenario has to anticipate player actions, and the longer the scenario is, the further into the future the scenario writer has to project the game events. Unless the scenario is endlessly filled with stuff like "make up something clever to happen here depending on how the PCs dealt with things so far" then someone writing a 20-session event-driven campaign for commercial publication pretty much has to assume the GM will railroad by nudging or nullifying in oredr to spit out a continuus stream of encounters. People with published scenarios need to learn to improvise if they want to get around this.
The inimitable Zak S. has written a massive 6200-word, laryngitis-fueled treatise on the different varieties of campaigns and their relation to the concept of the dreaded "railroad." Aside from being, I think, a landmark piece of RPG blogging in general, the post was of especial interest to me in relation to my ongoing Pendragon campaign.
The quote above pretty much sums up my feelings about the GPC. As one of the largest published scenarios ever, the book pretty much by default falls into the trap described in the quote. My big "zen moment" when using the GPC for my 2006-2008 Pendragon campaign was that this is, in effect, a feature of the book rather than a bug.
Here's an example from that campaign: the PCs were enemies of the Orkney clan, and in particular nurtured a particular dislike for Gawain. This culminated in a duel between the party's best fighter and Gawain, and in the first round of combat the PC in question laid Gawain out with a critical hit that took him down to single digit hit points.
At the time I breathed a sigh of relief--what if the PC had killed one of the main NPCs of the campaign around whom so many later events would revolve? The question bothered me so much, I posted it to a Pendragon mailing list I was on at the time. Someone--it might have been Greg Stafford, but I forget exactly--pointed out that had the PC killed Gawain, it would have actually been an opportunity for a great turn of events in the campaign. Arthur, impressed with the PC's prowess, grants him Gawain's former lands and responsibilities. Now all the "fun" things that were in store for Gawain are instead in store for the PC!
What the PC does with those events, or how he avoids them entirely, then becomes a major element of the game. In other words, as the quote from Zak's post states, improvising around programmed events on both a macro and micro level is the key to successfully running any published scenario, and this goes double for any scenario of great length. To take another quote from Zak's post:
My plan would be: overdo it. Every time the PCs make a decision, find a way to make the next event reflect it, even if you don't have to. This is harder than sticking to the plan and requires quick thinking, but it's also more fun for everybody involved.
This happened quite recently in my current Pendragon campaign, when Des came up with the idea to arrange a marriage between Countess Ellen and King Nanteleod. This was a possibility I hadn't considered, but as soon as she suggested it I was prepared to run with it all the way. The long-term ramifications of allying Salisbury to a powerful kingdom that will feature significantly in future events remain to be seen. Similarly, because her backup character married King Leodegrance, her lineage of future PCs will now be related by blood to Guenevere. That's obviously going to have ramifications on the campaign for generations--and ramifications resonating off of those ramifications and so forth.
All that being said, the fact that I'm running a single-player campaign does require a bit of railroading of the sort that Zak defines as "nursing"--basically making sure that there are sufficient NPCs around to bail out/help Herringdale if he gets himself in over his head. Back in the early days, when it was just me and my buddy Alex, I had to resort to similar measures, especially when I'd run AD&D. It's just the nature of running single-player campaigns, as most RPG systems are built with the assumption of multiple redundancies via a group of PCs; there will always be someone else around to haul an injured PC out of combat or cast the right spell or whatever. Take all that away and you've got situations where the villains are obliged to merely capture the PC they've rendered unconscious rather than kill them outright, or have a friendly centaur shaman cast stone to flesh on the hapless illusionist who was petrified in the middle of nowhere by a randomly-generated cockatrice. So it goes.
Like, I think, most GMs, I'm constantly evaluating my ratio of "story events vs. player freedom." I think it would be impossible for me to run a "pure" West Marches-style sandbox campaign. I (and for the most part my players) enjoy a framework of what Zak labels as a Clockwork Simulation to riff off of.
So for the most part, I like a having a degree of story-driven events (aka Unanticipatable Events), but what I most enjoy is encouraging player freedom within the context of ongoing events. As a GM, I like to be surprised and thrown for a loop. I like proactive players. And I like having to think on my feet as I improvise around scripted events. In other words, the railroad, if used wisely, need not be an enemy.