Make sure you are able to insert series resistors or ferrite beads (0402 size) in the ethernet lines connected to the phy.
I second that. Those ferrite beads can be a source of nasty resonances when combined with MLCC capacitors. A few ohms can safe your life at the test facility. (and make you look like a a genius.)
Keep in mind where this advice is coming from (I think). nctnico has worked with, LAN- or KSZ-something PHYs from Microchip, which emitted significant wideband noise not rejected by magnetics. Remember the magnetics are only rated to 100MHz or thereabouts, if that; and with fairly middling specs, at least for the minimum IEEE 802.3-whatever requirements they have to meet -- how much any given set exceeds the standard, they never tell, though. You can't expect balance (CMRR) at such frequencies (100s MHz, GHz), or filtering.
(There will generally be a notch in the transformer's transfer function due to the TLT construction, but we can only know here that it's above some minimum, and in lieu of filtering, there will be plenty of pass bands above that cutoff/breakup frequency, making it likely that harmonics go right on through.)
FBs with MLCCs is another matter. This probably refers to the AVCC bypasses around the PHY and magnetics. Indeed, they should be placed judiciously, and enough loss present in suitable locations (large ceramic + explicit resistor, tantalum, electrolytic, etc.) to dampen the system. I hope it goes without saying that large MLCCs (values that would be prone to resonate with FBs)
do not belong on the signal lines, which was the earlier context.
Note that midding-value FBs with small Cs (100s pF say) are generally safe, the resonance being in the 1s to 10s MHz, and the Q being small (<2?). Such values would a possible further improvement (as L section, reverse-L section, or pi network) to the mentioned problematic PHY outputs, with suitable values of course (cutoff above 100MHz; 10s Ω FB, 10s pF cap).
Tim