I'm running Phoenix 1.50 with a hacked pyopencl on Mac OS X 10.6.8 Snow Leopard on my Mac Pro 3,1 with an astonishingly overpriced 4870 card.
I get between 70 and 90 MH/sec depending on what other apps I'm running and how many Safari windows I've got open (normally 50+, with a 30" Apple LCD and secondary 18" Apple LCD). Certain Safari plugins hammer the GPU and eat 20 MH/s off the hashrate, whereas running in a alpha-blended (75%-translucent) Terminal window usually gets me 85-90 MH/sec.
This, according to the Hardware Comparison, is NOT BAD - especially given that the Mac isn't overclocked, the card isn't overclocked, and OS X doesn't get much love round these parts. I think Apple support OpenCL a **lot** more than the anti-Apple crowd make out... that's awesome performance from an old card, especially when I'm using the machine as my main workstation and won't tolerate any slowdowns.
I'm about to flash a PC version of a reference HD6870 which turned out to be pretty lame in Linux (the rest of my 6.5 GH/sec is a bunch of Linux frame rigs), which should make my Mac a little quicker in all graphics respects.
I'm not prepared to pay £370 or whatever for a Mac version of the 5870, indeed I'm not willing to pay current market rates for 5870s full stop - they're massively overpriced! However, the latest SL point release has drivers for 6870 Radeons so has been proven to work, even without flashing the card. Flashing the card with an EFI BIOS is only required so you can see a boot screen.
So to answer your question - I've had very usable results from Phoenix 1.50 on my Mac Pro and its ancient 4870. With that card, I had to use the poclbm kernel - phatk didn't work - but with a 5870 in your Mac you may have better results with phatk (or phatk2).
You need to know Mac OS X from the Unix side of things though. The only difference between getting Phoenix running on Mac OS X and the normal Linux method is that you need the modified pyopencl package, so python can use Apple's own OpenCL libraries (which are impressively quick, all things considered). I used standard Phoenix code, standard poclbm kernel, and didn't need to install the AMD drivers etc. because OS X already has them. It goes without saying that you'll need to have installed the Developer Tools on your Mac, since compilation is required, and also I'd advise that you install MacPorts because you may want to install a specific parallel implementation of Python.
However, it works, and it works well. The main difference is that AFAIK there are no overclocking tools for the Radeon cards under OS X. You have to stick with the Apple standard, unless you install a pre-overclocked (in BIOS) and then EFI-hacked PC card. I may try this, but I don't want hacked cards in my main production workstation...