I had a similar experience working with Jifty. I found the transparency on the database layer very good, and was able to get a lot of mileage out of that. When it came to the actual web code, however, I found Jifty getting in my way more than helping me out. Most of our bugs tracked back to a poorly documented user permission structure. After several days spent crawling through Jifty looking for what was set wrong, we eventually decided to strip out Jifty and build up a simpler system based just on raw Mason. Even factoring in the conversion time, it was a net gain in productivity.

YMMV of course, and it should be noted we were working on a fairly simple application where rolling our own interface from scratch was not that big of a deal.

I had a very similar experience regarding Jifty's user permission built-ins (and the docs which ostensibly describe them). Really, the whole project was a sequence of:
• Try to find FEATURE_X in the docs. On failure, skip next step.
• Try to extrapolate proper usage of FEATURE_X in REAL_LIFE_SITUATION_Y from the single provided line of sample code.
• Beat head against wall for several hours.
• Bug Jesse Vincent on IRC. Jesse provides the Right Way to do it in about ten seconds and 35 characters.
• Code it. The solution generally works and is elegant.
• 20 GOTO 10
I can easily imagine how keeping Mason and throwing the rest out could make your life a lot more pleasant, especially if you don't need AJAX/continuations/insulation from SQL/a pony/other features that Jifty provides out of the box.