I don't know that QED software at all, but considering the computer it seems to have been made for (36 bit GE), some information can be used for an educated guess:
but what is a Hollerith file?
Well, Hollerith can have, when it's about encoding, two meanings:
Punch card codes, as known from IBM punch cards. These are often associated with EBCDIC, but they are not, they are a 1..3 out of 12 encoding.
Especially in context of 36 bit machines, like the GE systems were, Hollerith code means a 6 bit per character encoding able to display all valid alphanumeric encodings used on cards (*1)
So my guess would be that a Hollerith file is one using 6 bit characters, independent of the source media.
GE lingo is a bit more specific, as it refers in manuals (like for the 225 reader, see p.12 *2) to 6 bit characters as BCD characters (*3), while Hollerith is reserved only for punch cards - a distinction not all users may have followed.
In addition, the term file, nowadays usually rather confined to mean some disk stored data, was back then a way more generic term more like today's "stream". A file could be/come from a paper tape, a card stack, a tape or any other means of data. Data that often never touched a disk storage at all. Like reading in a Hollerith data from card and outputting on a TTY. Both may have been called a file.
Is it related to the 9-vs-6 in the routine names?
Just a guess, but a 36 bit word can store 6 character using 6 bit Hollerith, or 4 ASCII characters by using 9 bits per character. It was general advisable to not let characters cross word boundaries, as that would complicate decoding. Files were of course usually character based. 6,7 or 8 per character, depending on system and encoding - or a punch character in case of cards :))
The names of these routines sound quite like it's about packing and unpacking such words, depending on what encoding is used. 4 characters per 36 bit when using (7bit) ASCII and 6 when using Hollerith.
This interpretation gets some support when looking at the GCOS Wiki-entry, where the section about storage mentioned the use of 4 ASCII or 6 BCD characters per word, while the QED page states that units of 4 words are used as basic building blocks for text storage.
*1 - What was valid was in part depending on each manufacturer, as each used variations of punch card codes for their own purpose - a bit like 8 bit home computers often botched ASCII to handle their additional graphics or diverging glyphs - think Commodore giving the pound sign a code position independent of hash (#).
*2 - To avoid confusion, don't look at p.13, as the GE-200 were 20 bit machines, so data a 20 bit word always held 3 right justified 6 bit characters :)) It was the only I could find showing the most likely encoding table.
This 20 bit word structure is, BTW, also the reason why all (classic) BASIC function have three letters names. Dartmouth BASIC was first implemented on a GE-235, with its 20 bit word able to pack three 6 bit characters. Thus all the now common names (almost canon, across many languages) like
TAN are due the wish to simplify the compiler by having to compare only a single machine word when looking up for a match. It also explains why Arcus TANgens isn't
*3 - The term BCD character might be a bit confusing, as it does not describe a 4 bit value (or two thereof), but two octal numbers used to spell out 6 bit codes.