1) Continuous operation in the 75-100 percent of nameplate power rating will generally deliver optimum efficiency. However, too much "extra' HP will result in a lower power factor (since the machine is running at less than nameplate loading), which will adversely affect power consumption and utility costs in the long run. As David mentioned - having a bit more than the nameplate value can be a good thing when process changes (such as colder-than-estimated gases) occur. Of course, the best thing to have done was to verify MAX power required at WORST CASE process conditions ... in which case, there's no need to add more overload capability.
2) Right now, the process requires some power throughput to achieve the end goal. Is that likely to change somewhere down the road - to a different speed or power rating? The relative cost of "new motor later with attendant in/out costs" should be weighed against "less efficient motor now costing money to run, but will work through life of process". This is more important for larger, more expensive machines of course. And as we all know, the costs associated with products and labor are always increasing over time!
3) I trust you've done the correct thing when looking at RMS rating of the process. But just as a reminder - look at ALL the loading conditions (including the peak and dip cycling on a reciprocating compressor application) and the length of time at which the motor is at different loads. Recall that for most machines, the thermal time constant of the winding is in the neighborhood of 5 minutes - which means if the "process" requires longer than that to complete a single "cycle", then the average has to be examined as a rolling window throughout the total process span. THEN pick the worst case from that, and use it as the motor rating.
I had an application for a steel mill once, where the end user was SURE the motor had to be rated for 6000 hp continuous (he had averaged the requirements over the entire 28 minute window required to process a single coil). However, looking at the actual loading cycle over the entire period, it became very obvious that the "average" loading for the first 5 minutes, when most of the work was being done to the material, was more like 10500 hp. For the last 15 minutes, the "average" was down around 4500 hp. If I had built his 6000 hp rating, it would have been irreparably damaged in the first few weeks of operation. So making a 9000 hp machine with a bit of extra thermal margin and stamping the NAMEPLATE at 6000 hp satisfied both the "make it safe" mentality of the engineer (me) and the "I got what I wanted" mentality of the user.