When you get into exceptionally small carbide bits the math does tend to change, and with 90 degree included angle (45 degree half angle) engraving bits especially you can run them very fast without issues - people recommend different optimal feed rates based on personal experience but they are all at least 2x and often 4x-5x or more than you would think over using larger bit calculations.
Remember that PPS isn't really a measure you should be using for stepper performance. It get complicated and involves a lot of things like voltage and inductance, but RPM is really what you should be looking at. Whether you are at 2000pps at full steps or 16000pps at eighth steps, the motor is still running at the same RPM which is more indicative of performance.
Secondly, remember that a stepper at a 24v supply might drop off to very little torque above, say, 20 RPM - while the same stepper driven by 48 volts might have more torque even at 40 RPM. At 72 volts, the stepper could keep the same torque well through 60 or 70 RPM.
In a related note, you likely need a HELL of a lot less torque than you think. There are calculations to figure out exactly what you need, but in effect if you can supply more torque than the amount of side loading it takes to snap your bit, the weak link in the chain isn't your steppers. Even 20 oz/in multiplied by 5 or 10 or 20 times by the mechanical advantage of the screws is still enough to do some seriously heavy work and or damage things. Don't worry about dropping below the 115/oz range - with the screws you mentioned that would mean around two hundred pounds of motive force at its lowest ebb, well beyond what you need or can use unless you plan on sitting on the worktable while its cutting...
"most machine only deform when your trying to push through material and that's where most of your backlash comes from right?"
Mechanical deflection and backlash are unrelated. Backlash is simply a problem with play in the screw/nut interface causing lost motion when reversing direction. (When the screw changes direction, it spins a little before taking up the slack and catching the screws opposing thread faces, causing the nut to "miss" some of the screws motion). This can happen in any size or rigidity of machine.
A machine flexing or vibrating under load is a bad thing, and an unrelated one to backlash. You are correct, however, in that you will have very little loading just milling PCB's and therefore mechanical deflection is not a big problem.
You likely need a system with far faster speeds at far less torque than you think, with a higher voltage supply and lower inductance motors. Trust me, for this stuff, the maximum 25,000 or 35,000 pulses that Mach delivers each second can actually be the limiting factor!