I don't know if it's true, but I've been told, by so-called experts, that performance takes a hit on a P25 Phase I system, when they switch to encrypted.
That was true with older platforms like Aegis, where audio was digitized and encrypted, then sent over an analog radio path. That's not how P25 encryption is applied.
Then there's the cost factor, and oh yea, so-called transparency in government. LOL!
Transparency in government does not require unencrypted radio communications. One popular argument against encryption states that the public loses oversight over government when police communications are encrypted, but encryption doesn't alter the facts that police communications are discoverable as evidence in court, and it's the grand jury that acts as civilian oversight over local government. Grand Juries can easily subpoena decrypted recordings to provide that necessary oversight.
My good friend is a fireman, and he said the P25 radios have enough problems when they're inside burning structures.
That's a common complaint. I look at it like this... (in much simplified terms)
Nyquist theorem states that in order to accurately sample an analog signal and digitize it, the sample rate must be twice the highest frequency to be sampled. I don't know, offhand, the sample rate used, but as an example for discussion, telephone transmission uses an 8 KHz sample rate to represent a 4 KHz audio channel. With 8 bit encoding, this results in a 64 kb data rate.
For P25, that raw sample produces a continuously changing digital word that, if it were to be directly modulated to RF, the bandwidth would be VERY very wide, in comparison to allowable channel bandwidths.
So, it undergoes compression via mathematical algorithms that essentially rebuilds the audio from an abbreviated digital instruction set, as opposed to transmitting a full digitized version of the audio.
A lot of fidelity gets lost in the process, because you've taken raw data that runs in several tens of KHz, and reduced it to 12.5 or 6.25 KHz occupied bandwidth. The algorithms are very good, and the complex modulation schemes used to apply it to RF are impressive, but what seems to be lost is voice recognition. It comes out sounding rather generic. Human speech is quite complex, communicating things via inflection and accentuating certain sounds, that simply doesn't translate to highly compressed digital audio. Therefore, it stands to reason, information contained in the original audio is lost in the process.
I've also noted in some tests I've observed that a rapid talker can seemingly "out talk" the algorithms, and the end result is received audio that sounds clear and noise free, but you have to ask for a retransmission because you just didn't understand it. It's an effect that's hard to describe and harder to quantify, but it's the only way I can explain what I've heard on strong, noise free signals, where asking the sender to talk just a bit slower makes a world of difference in intelligibility.
In a fire situation, add a high back ground noise level competing with the voice for the limited number of bits available for transmission, and intelligibility suffers even more.
I'm not at all surprised that the fire service frequently rejects P25 for fireground channels.