I would go for downbond for ground for shorter ground inductance. Although its true that adjacent bondwires carrying complementary currents will have mutual inductance as to reduce the effective self inductance of the pair, will your gnd and vdd really carry differential RF currents (at frequencies the inductance becomes noticable) ? what's the pcb arrangements for vdd and ground square ? I think its only safe to assume differential RF currents in things like differential VCO with equal bondwires to a common landing point/package pin.

Anyway you can do some simulations or hand calculations, as it'll depends on the lengths and spacings of the wires... assuming 25um diameter then 1mm down bound ground is ~ 1nH, and say vdd is 2mm (~2.2nH).
Two parallel 2mm tracks spaced at 0.4mm have a mutual ~0.6nH, so in hairpin mode the series inductance of gnd & vdd is 3.2nH (assuming IDEAL differential currents), i.e more than 3nH.

ACWWong's advice is bang on the money. Also use multiple down bonds for the GND's - pepper your pad ring with GND pads.

BTW, anyone know of a situation where GND bond inductance improves signal integrity? Not referring to amplifier source degenration here.

M.G.Rajan

The above statement about the L-degenerated LNA is wrong, since differential signals do not see this common-mode inductor.  The real impedance introduced by the ground bondwire is only for the single-ended L-degenerated LNA case.  To achieve input impedance matching for the differential LNA, on-chip inductors are needed.  Sorry for the confusion.  

frank