You want a prime number of stages to ensure that you
can't get a higher order oscillation (a possibility with, say,
a 15-stage RO - will oscillate indefinitely at 1/(15*tpd),
1/(5*tpd) or 1/(3*tpd) depending on how the "traveling
wave" initializes or is modified by switching noises.
Building in a good stiff decoupling on-chip is important
to getting ideal fosc measurements - if the supply is
collapsing, against your nH-range bondwire feed, the
oscillation frequency will be compromised or even
quenched / destabilized. You'd be well off to simulate
with these "external" parasitics, not just the close-in
extracted ones, with a reasonably inductive power
source.
You also want to rebuffer your output, I'd suggest with
a resistor divider or source follower to 50 ohms Zout and,
if practical, give this a separate supply and ground so as
not to perturb the RO with external load currents.