I've been playing around with this game for the past week or so, and found a radical route change for Bowser's palace that saves a small amount of time for the warpless run: (start watching at frame 34,600; the beginning is just the published run)
http://tom.seufert.googlepages.com/lordtomsmb3wip1.fcm
Hopefully you guys like it; I certainly expect some of you will be a bit surprised at the route I take, and hopefully find it more entertaining, too. ;)
This WIP is *not* fully optimized, but saves 11 frames, judged by the frame on which Bowser 1st appears (frame 36377 in the published run, frame 36366 in my wip). Start watching at around frame 34k...
I tried many strategies to make the wall jump faster, including duck-jumping from mid-air when mario 1st appears, but was about 4 pixels shy no matter what I did. I know I'm missing corner boosts (intentionally left out to make proof of concept more quickly), and the final screen before Bowser can likely be several frames faster.
I'd like to submit an improvement based on this once I get it optimized; if any more experience mario tas'ers have advice, I'd love to hear it via pm, the forums, etc.