The biggest problem with IMU's which are out there for the hobby market is the accelerometer. 8G's is not enough for most rockets, so now you need two accelerometers - one for the fine details (fine res, low G) and one for the big G's (coarse res, high G). You also need a GPS which can handle the G forces - not physically but in software. With the speeds involved in a project like this, and the boost phase a hobby system running on an Atmega (eg: Arduino based software) simply wont cut it with the 50-200hz you can get out of it with highly optimised software - 8 bit systems also severely impede speed due to boxing and unboxing for larger values. This means you now need a mid to high speed ARM or ARM Cortex - a decent R4 or M3 will get you 120-180mhz which will get you 500-1500hz depending on the sensors you're using.
Obviously, hobbyists are more than capable of using these parts, however the "arduino generation" are going to struggle to create an optimised solution. This is where costs come in as you need developers who are extremely competent with optimisations at a low level, as well as having a firm understanding of the complex maths and weightings involved. Software for working on ARM which will give you the debugging and development capabilities starts at $1500 or so, and an optimised compiler (Keil) can easily set you back $10k. You also need an electronic designer who can optimally route and simulate the board so you're not getting noise introduced and who can manage the thermal profile of the board so drift is not artificially introduced. Software like this (Altium) starts at $5-6k. Whilst you could do this with a package like Eagle, and an open source tool chain for software you're not going to have an optimal solution - or if you do it takes months and months longer.
As far as GPS goes, i've worked extensively with SkyTraq who make the venus chips. They are very responsive and usually pretty happy to implement custom software. They already have a firmware version with less filtering which can handle rapid accelerations and less-linear trajectories. They also implement ITAR as an AND rather than an OR - meaning over 60,000ft AND over mach whatever - i dont remember the exact speed off the top of my head on a friday morning! They also do 20hz GPS data and 1.2m HDOP, not just 1 like, say, SIRF lol.
Sensor fusion like this is not easy, but its hardly the hardest thing i've ever worked on. The biggest issue here is time, and when time is money a good IMU will cost you thousands as others have said.
