This was an oversight when checking the regression bitmaps. Go to https://ghostscript.com/~regression/release__gs/compare71.html search for '6245:' At the bottom left corner of the first bitmap, just above the label "Page=1 PageSize=2480x3508 Res=600 TopLeft=(125,2046) W=516 H=960 388 X 2544 Y" there is light blue rectangular area containing two darker blue horizontal bars, and three darker blue circles. An erroneous lighter rectangular area can be seen intersecting the top of the circles and extending to either side. I can reproduce this on the current HEAD (SHA ce437b88df500a184ae2ca15d67a019f57a11d20) using: gswin32 -sDEVICE=psdcmykog -sOutputFile=out.psd 575_-_images_missing_blend_modes.pdf Adding -r300 makes the problem go away so I suspect this may be clist related.
Both 300 and 600 are banded, the default (600) uses bandheight 28 and with -r300 the bandheight is 56. Changing the -dBnadHeight to 30 moves the difference, slightly, so I agree that it is banding related. First step, try and isolate this mess down to something simpler.
This appears to have been fixed sometime during the course of the last 4.5 years.