As promised--here is a Windows batch file which uses cpdf to keep the original bitmaps but add a text layer with k2pdfopt's OCR. Should be easy to convert to linux.
Code:
rem
rem Step 1. Do the OCR
rem Typically set dpi to at least 300 for best results
rem -ocrd p sets detection at the page level. This means
rem that the Tesseract algorithm will be used to
rem find text on the page rather than the k2pdfopt
rem algorithm.
rem You may also wish to add -g, -cmax, or -s option
rem adjustments to improve the bitmap contrast or
rem sharpness and resulting OCR quality.
rem
k2pdfopt -mode copy -dpi 300 -ocr t -ocrd p src.pdf -o temp1.pdf
rem
rem Step 2. Replace the bitmap in the result with a very low density,
rem low res bitmap (which will later be ignored / made invisible),
rem but keep the text layer.
rem
k2pdfopt -mode copy -dpi 5 -bpc 1 -g 100 -cmax -100 -s- temp1.pdf -o temp2.pdf
del /q temp1.pdf
rem
rem Step 3. Pair the text layer in temp2.pdf with the bitmaps in src.pdf.
rem Put the result in src_searchable.pdf.
rem
cpdf -draft temp2.pdf -o temp3.pdf
del /q temp2.pdf
cpdf -combine-pages src.pdf temp3.pdf -o src_searchable.pdf
del /q temp3.pdf