I wonder how much would need changing in NR when moving to quadruple precision. I think everything except the special functions is fairly precision-independent, right? Of the special functions, some have defined convergence tolerances which 'semi-automatically' can be adjusted, whereas others are based on some fixed-precision approximations. From what I can see all these 'fixed-precision' approximations are single-precision, so that the dp routines seem readily upgradeable to qp? am i missing anything subtle before I take the plunge?

I think you are about right -- but no guarantees!
Cheers,
Bill P.