So if I understand the problem you have, it is the same as one that I complained about also. This issue is independent of how you test the altimeter, either real life or in a vacuum chamber.
To restate the problem: the altitude beeped out by the altimeter does not correlate to the data downloaded from that flight onto a PC and plotted in mDACS. The peak altitude observed for the SAME flight is different.
This is the case for every real flight of my RRC3 as well, so it's not a vacuum chamber effect.
I contacted Jim Amos at missile works about this, but unfortunately I can't find the e-mail exchange about it.
If I remember the gist of the explanation correctly, it's that the time history data saved to the internal memory are processed by the mDACS software in a different way than the live altitude calculations the RRC3 performs. The beeped out altitude is based on the live RRC3 calcs, but the mDACS data on the PC is (obviously) processed by the PC. This answer didn't completely make sense to me (EDIT: I thought you'd just use the exact same atmosphere model for both and crunch the same data and get the same result, but perhaps it's due to math limitations of the embedded processor versus a PC), and I may be recalling it incorrectly, but I just gave up on trying to understand it and decided to just not worry about it.
I also noticed another discrepancy related to the arming altitude. I noticed on the PC graph instances where the landing altitude is lower than the launch altitude. I don't know that this is related to the problem noted in the original post, but I found it to be moderately annoying as well. This was explained as being due to how much data can be stored "pre-trigger", where the trigger event is the rocket hitting the arming altitude. The system only saves X number of data points prior to the launch detection event (tied to the arming altitude). A rocket that accelerates slowly and takes a while to hit the arming altitude may experience loss of the early on-the-pad pressure data points. As a result, the graph is treating the first data point found as the launch pad elevation, when in reality the rocket was already in flight at that time. It isn't anything super troubling, but it just triggers my OCD a bit when post processing flight info.