[pirl8]

Originally Posted by JSWolf

Do you know how much larger the eBook would actually be with &shy added to every place possible that a hyphen could be?

Yes, of course I know.

The html part is rougly twice the original, estimated with a prefix/suffix of 2 characters, which gives the highest amount of hyphenation, considering an average of 2.5 characters per syllable and estimating 5 characters to store the "­" entity (which gets actually easily compressed).

Spoiler:

Just put the HTML files in a directory and run this script to have an estimate

#!/bin/bash
for F in *.html *.xhtml
  do lynx --assume-charset UTF8 --dump $F | 
  tr '[:alpha:]' x | 
  tr -cs 'x' '[\n*]' 
done | 
sort | 
uniq -c | 
sort -n | 
grep xxxx | 
awk 'BEGIN { \
        SL=2.5; \
        TOT=0; \
        SHY=5; \
    } \
    { 
        print $1 " instances of " length($2) " character long words" ; \
        TOT += $1 * (length($2) / SL + 0);
    } \
    END {
        print "Estimated text increase: " SHY*TOT \
    } ' 
echo -n "Total text: "
du -c -b *.html | tail -1 | awk '{print $1}'

There are no memory concerns (embedding fonts - even with subsampling - or adding oversized images for a best view with KPW which has an higher resolution uses much more memory and images cannot be compressed like text) and - at the moment - this is the ONLY way to have hyphenation on AZW3 files on e-ink Kindles.