This feature request really comes down to a request for an additional standard way for a user to provide the XSLT processor with nonstandard language-specific options.

The specific issues for "traditional Hebrew" are:

Sometimes numbers are printed with additional marks to indicate that they are numbers, sometimes they aren't.  The specification uses both conventions, once in the example for dates, once in the example for numbering.  The types of additional marks also change.  In modern texts, numbers are sometimes marked with a geresh (1=א׳) following the number, and sometimes with a gershayim (21=כ״א); In archaic texts, overdots are sometimes used to indicate that the value is numeric and not a word (21=כׄאׄ).

When the number is represented as words, it could be masculine or feminine, in both ordinal and cardinal forms.  There's currently no way to specify masculine or feminine for cardinal forms.

Also, there are two conventions for how to specify a number in words.  The modern convention (the equivalent of representing 1234 as "one thousand two hundred thirty four") and an archaic convention ("four and thirty and two hundred and one thousand").

It's probably beyond the scope of XSLT to define all possible options for all possible languages, but, in the interests of interoperability between processors, ideally, the set of options for each language could be expanded in some way involving community-consensus, perhaps by a list of non-normative best practices for each language.

A model for how to do it might be in the @ordinal attribute of xsl:number (XSLT 2.0, sec 12.3):
"For inflected languages that vary the ending of the word, the preferred approach is to indicate the required ending, preceded by a hyphen: for example in German, appropriate values are -e, -er, -es, -en."

Similarly, a new @options attribute to indicate language-specific options could be added to the spec.

The XSL Working Group approved this response at its November 18 2010 Teleconference.

Your comment highlights a number of the difficulties in supporting
the wide variety of conventions used over time in many languages.

Some of these cannot even be represented in a "linear sequence" of
Unicode characters - such as a "titlo" that is "stretched" to cover
all the characters in an old slavic number.

For most of the variations that you list the intention was coverage
by (just before the description of "grouping-separator")

  Note:
  Implementations may use extension attributes on xsl:number
  to provide additional control over the way in which numbers
  are formatted.

Do you think this should be augmented and clarified to get the point
across? One could - if you would volunteer to organze a "community"
design effort - add a specific example designed to fully cover a
particular language. I feel that a SINGLE attribute will be too
limiting in the general case so a "standard" @options is probably
not the way to go. The namespace of the extension attributes should
probably clearly identify the language and "authority" of the
definition.

Could you provide examples of where the cardinal form varies based on
e.g. gender? Preferably in more than one language.

Since we have not heard back from the commenter for over a year we are closing this bug. If you feel this is in error please supply requested examples and reopen.  Thank you!