An FDR isn't a very good model, because it implies no phase shift, and imaginary properties (like non-causality). Of course, any real, physical, two-terminal "FDR" has to have reactance, so it's very easily approximated with a finite string of RLC units.
Most SPICE models you will find, are a single RLC, or RLC + DCR. This works okay over a few decades, but not usually to great accuracy over the entire specified range of the FB. Tweaking a model to match the published curves isn't hard to do.
Anyway, even such a simple model answers your question: yes, it adds noise, and only at the frequencies where the resistance component is significant. If the FB is chosen so the resistance peak occurs at frequencies outside your passband, then system noise is unaffected.
Another way to think of it: if you can replace your series resistor with an R || L, such that the F = R / (2*pi*L) cutoff frequency is above the top passband edge, while still achieving the desired damping, then you're set. Or to put it another way, you've constructed your own ferrite bead, but with more accurate components.
This can only work, if the series resonance you're trying to dampen, falls at a frequency above the passband. Which means the lead lengths must be kept short enough for this to be true.
Mind that real ferrite beads also saturate under DC current bias, and have interesting AC bias properties (the inductance rises for small (nonzero) amplitudes, then falls as amplitude approaches saturation). This isn't a problem for small signal applications, but precludes ferrite beads for high current filtering.
Tim