First, the magnitude of the New Madrid earthquakes is continually being revised downward. Now, we think the biggest of them was in the mid-7 range, so it is not so puzzling that the fault lengths are so short. Mike is right, the midwest rock is probably more intact than on the west coast, but this shows up more in less attenuation of the waves in the East than a difference in the magnitude-length ratio.

Second, there is a debate in the community between those who think faults are dominated by a characteristic behavior (every big quake on some section is an 8, for example) and those who simply expect a Gutenberg-Richter distribution (a mix of all magnitudes, with more little events, of course).

Third, faults are not simply divided into segments. Evidence points to large, multi-segment earthquakes causing more slip on a given segment than do single-segment earthquakes, although evidence is slim.

Fourth, slip is observed to be highly irregular across a fault in a large earthquake. So 1906, even if it broke the same fault as Loma Prieta, which is not certain, may not have had much slip there. Last time I checked, the geodetic (Marshall and Stein?) model for 1906 looked quite different than the seismic model (Wald?), so we don't know the distribution well for a quake that old and ill-recorded.

Last, the comparison between the shaking of a local M7 and a more distant M8 needs to mention frequency content. The high-frequency shaking may be greater for the local M7, but the long-period shaking, the kind that damages large building most, would be almost invariably greater for the M8 event.