Ben's comments are right on.
The 7030/6040E has a "motherboard" (7030) and DAQ "daughterboard" (6040E).
The DAQ board in use (6040E vs. 6052E) makes no difference in terms of maximum PID or other control loop rate. This is controlled by the processor the code runs on. In the case of the 6040E, your code is running on the 7030 which as a 133 MHZ 486 processor. As documented in the below link, the fastest you can expect to see on a single channel is just over 1 kHz, as you are seeing. A high-end PXI RT controller would be the PXI-8176, which has a Pentium III 1.2 GHz processor. This type of controller with a PXI-6052E or PXI-6040E or other E-Series DAQ device would likely be able to run your code (which sounds to not be the problem) at 40 kHz or faster, as shown in the second link below.
The fact that your code runs at 3 kHz on a Windows machine running a 900 MHz processor is not surprising, but what you will find with Windows is that this 3 kHz rate is not consistant. It might even run at 3 kHz 99% of the time, but every now and then Windows will take over your system and cause a loop to take a full ms or more to execute. That, as Ben mentioned, is what RT is for: to guarantee a small upper bound on this variation in loop time. In the case of your 7030, you can only achieve 1000 Hz, however, each loop iteration should be 1 ms with variation of only a few microseconds. In Windows you will find very large (ms-or-larger) variation in loop cycle times (this variation is also known as "jitter") that are generally unbounded in size.
