The merged event list should not be used for spectral analysis, since it does not contain sufficient information to generate correct response files. The recommended technique for the spectral analysis case is to generate separate PHA, RMF, and ARF files for each observation and to analyze them simultaneously.This requirement is simple to understand--all the tools that produce calibration data products (e.g. mkarf, mkacisrmf, PSF generating tools like MARX) need to know where on the detector the source lies. To the extent that the ObsIDs that you reproject and merge are dissimilar in their pointing position and roll angle, the notion of a single position on the detector becomes undefined. Thus, judgment on your part is required. Some projects that have the luxury of a single pointing with fairly consistent roll angles among all ObsIDs have used a strategy of combining the data to make it look like one ObsID (with long "bad time intervals").
Small shifts and roll angle variations among merged ObsIDs are expected to induce only small distortions to PSF models and RMFs, because these characteristics vary smoothly with position on the detector. The ARF also depends in part on smoothly varying quantities (vignetting and detector QE). However, the ARF also depends on a quantity with high spatial frequency--the dither pattern of the CCD edges and CCD bad columns, projected onto the sky--as can be seen in the exposure map. Thus, a source extracted from reprojected and merged data will suffer significant calibration errors if in reality it falls on very different features in the single-ObsID exposure maps.
Observers sometimes wish to tweak the astrometry of the ACIS data to better match an astrometric reference. We recommend that you choose a method of astrometry adjustment that will not change (within CIAO) the inferred location of events on the detector. The only astrometry tweaking method that we feel comfortable with is changing the RA and DEC columns in the aspect time series (i.e. ``fixing'' the aspect solution itself), and then regenerating the event list with acis_process_events or with reproject_events.