Hard to answer this question without getting into fairly arcane technical details but I'll have a go.
The MDAC is using ESS9018 which is a sigma-delta based chip using noise-shaping to get the figures looking good. Noise shaping has various technical issues associated with it, many of which ESS have solved (compared to the other players in the field like ADI, TI/BB) but clearly not all. In particular the chip still exhibits shifts in the noise floor at particular output levels. Its also very hard to get enough subjective dynamics out due to the extremely high switching frequencies used in the on-chip DAC.
Metrum uses a ladder-type DAC and runs it at the lowest frequency (assuming you don't use your computer to oversample). Thus dynamics will tend to be better, also timbral accuracy. However the measurements relative to any ESS-based converter will suck - THD+N in particular and the FR will droop due to the NOS intrinsic roll-off.
Incidentally you're right there's something adrift with the Metrum's digital input - the digital ground beyond the isolation trafo goes into the ground fill.This induces HF noise currents into the sensitive analog ground and is a very common issue with commercial DACs, akin to the 'pin1 problem' with professional XLR connected kit. For anyone with the guts to mod, this can be quite simply fixed up. Just re-route the 0V side of the S/PDIF input with a dedicated ground wire back to the PSU input.