Hi Rob,
I don't know if I can explain the Tf value differences either, but here's a run-down of the Alliance 2690/5 injector in contrast to what I recall about the Agilent's: (would help if I knew if the binary systems you speak of were 1100 or 1200 systems.)
Side-port needle. (The Aglient has a straight-thru needle.)
Seal Wash and Needle Wash. (Agilent has only a Needle Wash, washing mechanisms differ between the two vendors.)
Flow-Through Needle Design, sample loop is integrated into the eluent flow path. (Can't recall at the moment, but seem to remember that the quaternary gradient and binary gradient LCs Agilent sells may have either Flow-Through Needles or a Fixed Loop design where the loop is separate from the eluent flow path.)
The only guess I have is to check into the nature of the needle wash solvents used on both systems...as well as to understand exactly what is done when the needle is washed by the autosampler for both types of LC. This may be the source of the problem...or differences in the rise times for each instrument...longer rise times can be manifest in larger values of Tf, though your case would push this guess to an extreme such that it isn't that likely to be the cause.
Otherwise, unless the Agilent and/or Waters LCs were modified, another thing would be that the capillary tubing ID on the Alliance systems would be larger than those for the Agilents, 0.009" versus 0.007" in the standard configurations for each system.