Dear Karoline,
For 204 sequences, 0.1 is probably too liberal. Use 0.05, and also take a look at Multimedia File Viewing and Clickable Links are available for Registered Members only!! You need to
Regarding different results from different methods there could be several scenarios: if one of the methods is significant (0.02) and the other is borderline (e.g. 0.06), then they aren't really disagreeing; the difficult cases if when the methods give diametrically opposed results (which should be rare for large data sets).
In our experience, FEL is generally more accurate than SLAC sets, so I would go with it, and report SLAC results as a backup.
Global dN/dS ratio is fairly robust to the estimation method, so I would report the value from the SLAC page (with 95% confidence interval).
Rate estimation does imply a molecular clock. HyPhy can do it, so can TipDate and BEAST from the (former) members of the Oxford Evolutionary group. Take a look at DatedTipsMolecularClock analysis in HyPhy (it is under Molecular Clock in Standard Analyses).
You should be able to find strongly selected sites in 19 sequences; run all three methods on it; REL tends to have more power for this size.
Cheers,
Sergei