They did get it nicely right on the RCD855. If you're running the SAA7220 from its own crystal then, yes, you're going to run into all sorts of problems with I2S very audible data stream corruption. In Philips based players, the 7220 is generally used as the MCLK source for the player (although the 7210 also has the capability but I've never seen it used).
SCK (variously referred to as XSYS and SCK in various datasheets) needs to be 256 x I2S BCLK, so 11.2896MHz for 44.1kHz, 12.288MHZ for 48kHz - the highest you can run the SAA7220/TDA1541, 4 x oversampling. As moffy says, if you can't slave the Pi to MCLK then you will need to generate a synchronized MCLK from the Pi's BCLK output.
Irrc, you shouldn't need to worry about the subcode inputs, they are there to support the optional S/PDIF output. You will need to tie the error flag, mute and atten inputs correctly though.
Note that the SAA7220 can be a noisy chip. Good local regulation and decoupling are needed. Supply noise can also make the on-chip crystal oscillator circuit more jittery too. The TDA1541 also has an SCK input which is used to (optionally?) reclock its I2S data input for reduced jitter. This was abandoned on the TDA1541A as unnecessary (at the same time, the DEM clock oscillator capacitor was moved from on-chip to external for reduced jitter).
Be careful where you source your SAA7220s from. The ones from China (if they are actually filters) may be re-marked SAA7220PCs. The optimal matchings are SAA7220PA for the TDA1541, SAA7220PB for the TDA1541A (which doesn't include the offset that is needed to make the TDA1541 perform optimally and also has better filter ROM coefficients). The SAA7220PC was designed for early digital receiver use at 32kHz and is not optimal for normal usage.