If it's dissipating power with absolutely no load, then something is very wrong. I've never used that specific topology before, but it's likely that at no load the duty cycle is not going to zero (which it should with a decent controller). Probe pins 1 and 8 with no load and see what the duty cycle is.
I am assuming it gets hot below 7.5V input. If that's the case, looks to me like you may not be fully saturating the PNP switch. Not surprising, given you are saying 5V and the app note says 7.5V as a MINIMUM. With a low VCC you are not gong to drive Q2 hard enough with that latch, no matter what you do with those external resistors. The timing cap will be out of whack as well. if you have a scope, you'll see what's going on vs guessing.
The lower operating voltage will affect efficiency, but at no load it shouldn't be dissipating significant power regardless, so there's something else up too. Unless the chip simply can't oscillate correctly without a higher voltage (which is possible).
Also I'm surprised you can still find chips that recommend BJTs in there app notes...