It gets tricky using my method for comparison to other equipment from other manufacturers, simply because very few provide the data. For instance, there isn't the necessary data for the specific Heatercore radiators to allow me to compare a theoretical PA based system vs a theoretical Heatercore based system.
Have a look here:
http://www.over-clock.co.uk/ivb/index.p ... opic=20277
Reason why I don't work in CPU / GPU temps and instead work in coolant temps is that no-one has done consistent testing of all CPU and GPU blocks on a single testrig, and even if they had, a single testrig can't represent differing Die sizes / Heatfluxes. You'd have to have several testrigs, one producing data that's useable for quadcore, the other producing data for dualcore, doubled up - you'd need one of each for intel, one of each for AMD, as their die sizes differ and thus so does the heatflux etc. This means c/w figures from different sources can't be used simultaneously in one equation. C/W figures calculated for a waterblock using a dualcore processor can't be used to predict performance of a quadcore processor... and vice versa... and data derived from an AMD cpu as a heatsource can't be used for an Intel CPU... etc etc... so you have to find someone who has tested the same block with the right type of CPU. It all gets tricky. This means that really it's much easier to base it off known radiator thermal dissipation.
There are some brands where one can predict performance - eg: Swiftech. If you have a loop consisting of all Swiftech blocks then their C/W data tends to be useable, as long as you have a processor matching the characteristics of their c/w charts (ie: they have a chart for singledie cpus, and one for dual die cpus - make sure you use the right charts).
EG: To work out CPU temps...
ambient + ([radiator c/w]*[total_loop_heatload]) + ([cpu_wb_c/w]*[cpu_heatload])
To work out GPU temps...
ambient + ([radiator c/w]*[total_loop_heatload]) + ([gpu_wb_c/w]*[gpu_heatload])
ambient = temp at radiator air inlet.
The bolded section is this: ambient + differential.
radiatorc/w * total loop heatload = differential.
This calc usually done at full load. You can use 50w to approximate an idle graphics card if you want CPU temps with GPU idle... and the figures are available for CPU Idle Load if you want CPU idle with GPU at load, or indeed just predicted temps for both at idle. (Refer to Intel Whitepapers for their CPUs for their idle loads).
Armed with that info, you can now find out what CPU and GPU temps would be produced and deduce what the "ideal" max tolerable differential would be.
Bear in mind, all stated power consumption / heatloads out there tend to be OVER calculated - ie: they are above the actual real-life figures. If you calculate as above for your setup, you should always have some headroom as a result.
Calcing the temp of air at exit of the rad... again, tricky. Yes it COULD be calculated as a c/w figure could be produced... but someone would have to sit and produce the data to get that c/w figure, and it would have to be done for EVERY fan, with each fan being tested across a wide range of heatload and flowrate - a VERY longwinded task. I'll state now tho, everyone I know, myself included, all suck air into the case thru the rad - ie: mount the rad on chassis air inlets / use it as an air inlet, which means it's heated air is dumped into the chassis and exhausted out the back. I've never had any issues with case temps being over the odds... nor has anyone else I know who runs a similar config.
Max water temps - I personally like to keep 40 deg C as the very highest... much above this can cause tubing to soften etc, which runs the risk of the tubing collapsing, and in setups where clamps aren't used (eg: 7/16 ID over 1/2" barbs) could result in tubing sliding off the fittings.