ST and TI have 5V ARMs mainly for automotive use. But , in general, most of the common parts are 3v.

Power consumption is a contentious issue. One would think for a small power conscious application one would use an 8 bitter (or 16bitter for msp430). Nevertheless, for performance vs power consumption, the ARMs seem to do pretty well.

Free tools - there's WinARM and Yagarto for starters. With the ARM processors, code lives in the same address space as data, so there's no special tricks for the compiler to perform (like with 8051 and AVR) so the gcc compiler is quite mature and well tested.

Luminary Micro have some great evaluation boards, but the downside is their timer peripheral has some shortcomings - this may or may not be aan issue in many cases but for me having no prescaler on a 16bit timer clocked at 50MHz when doing input capture is a bot of a shortcoming. Also only 2k receive buffer for their ethernet parts can introduce some problems.

As for what processor to choose? Usually for me its the peripherals, price and availability. with the ARMs, there's plenty to choose from.