100 ppm (parts per million) accuracy in crystals is fairly commonplace these days, but if you have one unit that is 100 ppm slow and another that is 100 ppm fast, there is a 200 ppm rate skew between the two crystals. If you take 200 x 10^-6 times 3600 seconds in an hour, you get a total of 720 milliseconds per hour of timing skew between those two crystals.
If you wish to match one recorder to another, you can add capacitive loading to the crystal in the faster unit to slow it down to match the other. You'd connect a small (perhaps 20 pF low loss capacitor) from each lead of the crystal to the ground used by the active elements in the crystal oscillator circuit. To get it exactly, you'd also have to include a trimmer cap so that you could set the capacitance precisely. That will get you close as far as matching crystal rates at the same temperature, but they also vary over temperature and the use of capacitive trimming will affect that temperature dependency. So, you'd need a different value of capacitance at each temperature.
Also, it's not possible to pull most crystals by 200 ppm with capacitive loading, so to match two units, you have to start with matched crystals and trim out whatever small difference there is with your trimmer cap. In that case, you may not even need the fixed value caps mentioned in the paragraph above. You'd just need the trimmer cap from one lead of the crystal to ground in the recorder that has the faster sample rate of the two recorders.