I show about .3W dissipation on the differential transistors, but bear in mind that the current source (Q3) will be running twice the current, hence about .6W. Speaking purely for myself, I prefer not to run a TO-220 case over about a half watt without a heatsink, so I'd stick one on the current source. One of those little three, four, or five-fin heatsinks will be more than enough, and they're cheap. Whether you run one on the gain transistors is up to you. They'll be warm to the touch, but not hot.
As for the outputs, of those two I'd say go with the IRF640, purely because it gives you a little better SOA. The IRF540 would work fine, and might arguably sound a little better, but you'd be running closer to its limits, heat-wise.
I used IRF644s in my Aleph 2s--also a TO-220 case--but once I got the water-cooled dingus going, heat wasn't a problem for me. You'll have to judge these matters according to what heatsinks you can find.
cp642's point about the TO-247 case devices is valid. They give better heat transfer, but are more expensive. Your choice.
Grey