The D104's response peaks between 2Khz to 3Khz which it why it’s the ultimate mic for what we do. The grid resistor should be 1M min to 10M ohms max. Some people like to modify the old gear that came with 1M resistors with a 2.2M resistor. Using a 4.7M or even a 10M ohm grid resistor works better, but you can’t go higher than 10M. Usually if you go higher than 2.2M you need to also change the cathode resistor while checking the voltages and current on the tube.
That 10M maximum has to do with the rig, not the mic. If you use too much resistance in the grid circuit the stage becomes unstable and the plate current may wander, as a minuscule grid current adds an additional bias voltage to the tube. In the early days of radio, this was often done intentionally to provide "grid leak" bias, but proved to be too unstable for mass produced equipment. Sometimes the first amplifier stage in a low-level preamp may use "contact bias", with no cathode resistor or grid bias resistor at all. I seem to recall that circuit in some of the speech amplifiers described in mid-50s Handbooks. The grid just floats, and the negative charge that builds up on the grid it what biases the tube. BTW, it is a misnomer to call the grid resistor of a class-A amplifier stage a "grid
leak resistor" unless grid leak bias is actually used to bias the circuit.
For most tubes, the RCA Receiving Tube Manual actually recommends something like a maximum of 0.5 megohms of resistance from grid to ground, but this would be too low for a crystal microphone. When you go above 5 megohms or so, it may be necessary to select tubes for the first stage to find one that is stable. The way I do it is, with no signal input to the grid, measure the voltage across the cathode resistor. Now short the grid to ground and measure again. Find the tube that shows the least cathode voltage change, thus the least plate current change, from the shorted to unshorted condition. My push-pull preamp uses 10M from each grid to ground. Instead of a 12AX7, I use a pair of 6F5s, which have nearly identical characteristics to one section of the 'AX7, but the single triode allows individual selection of each tube in the first stage. The push-pull configuration puts a total load of 20 megs across the mic. There is no need to use a grid blocking capacitor at the crystal mic input, since the crystal mic itself forms a capacitor and should measure infinite resistance across its output terminals.
If the mic sounds too bassy with the high grid resistance, that means the speech amplifier has insufficient high frequency response. You need to do something to boost the highs, not cut back the lows with less grid resistance. I use a simple method to generate a "presence rise", which consists of a specifically calculated low value of cathode bypass capacitance based on the time-constant formula. This low bypass capacitance causes the cathode to appear bypassed at higher frequencies, but at lower frequencies it acts as an unbypassed cathode resistor, and the degenerative feedback lowers the gain of the tube. Using two tubes in cascade with the special cathode bypass gives me about 10 dB of boost at the high frequency plateau beginning at about 2 kc. The rising response begins at about 800~ and rises steadily until it reaches that plateau. This presence rise is in addition to the resonant peak of the D-104. By limiting the peak to the upper midrange, beyond 2K, this doesn't make the audio sound tinny as long as there is sufficient bass response to balance the treble.