Baud with microcontrollers

My homebrew WX-1 weather station, which transmits data directly on 144.800MHz APRS using a PIC based packet modulator and a 10mW VHF transmitter module, is decoded by my VX-8GR and my Kenwood TM-D710 but not my TH-D72 or my homebrew TNC. This is annoying. A week ago I looked at the packet modulation from various devices and found that the tone frequencies of the WX-1 were about 50Hz too high. So I thought I would try to fix it.

The PIC source code for the weather station is available for download. The program code that generates the packet modulation is a bit beyond me, but I think it works by executing a loop and pushing binary values out of 4 ports which are connected to 1K0, 2K0, 3K9 and 5K1 value resistors in such a way as to produce a stepwise approximation of a sine wave. It seemed to me that to lower the tone frequencies I needed to slow the loop down a tad. After a bit of trial and error inserting a nop (no-operation) instruction in various likely-looking places I managed to get the tones nicely symmetrically positioned around the 1200Hz/2200Hz nominal frequencies for 1200baud packet. But the TH-D72 and homebrew TNC still refused to decode any packets!

Wondering what to try next, it occurred to me that the tone frequencies were not the only parameters of a packet signal. There is also the baud rate. I also remembered that the MixW sound card software prints out the measured baud rate next to each decoded packet. So I hooked MixW up to a transceiver and sound card and gave it a selection of signals to decode. I found that whilst my two Kenwood transceivers and the Yaesu VX-8GR all measured 1200 or 1199baud, the WX-1 recorded a value of 1208 baud. That had to be the cause of the problem.

Unfortunately I can't find a solution. I thought that slowing down the PIC processor's clock might make the necessary adjustment, so borrowing an idea from a PIC frequency counter circuit I replaced one of the fixed capacitors on the oscillator crystal with a variable one. This made a whole 1 baud of difference! Clearly that approach isn't going to get me anywhere unless I get myself a 19.867MHz crystal. If I'd done the math in the first place I'd have realized I wasn't going to pull the oscillator that far with a trimmer.

The solution, if there is one, has to be in the code. But I don't understand it nor can I find any comments that would point to a routine or value that affects the baud rate. Give me a circuit with discrete components any day. If this is the future of electronic experimentation I don't think there's a place for me in it.