Note: I have no personal experience with this calculator. Also, a lot of this has appeared already in comments.
References: the main HP 9100A/B page and the programming page at hpmuseum.org.
The 9100A/B worked with floating-point registers that had 10 visible decimal digits, two guard digits, and a two-digit exponent. This was represented internally by 14 BCD nibbles. (And, I guess, two additional sign bits for the mantissa and exponent.)
The 9100A had 16 storage registers (not counting the RPN stack) which could be used for either code or data. To maximize the amount of data space, of course the program counter would advance through a single register first before moving on to the next, so the low "digit" of the PC was effectively base 14. For some reason that isn't clear to me (maybe a limitation of the magnetic storage cards?) you could only use the first 14 of the registers for code, so the high digit of the PC was also base 14.
That was the only use of base 14. Presumably the PC was stored as two nibbles internally with some binary-coded-tetradecimal logic to increment it. Branch targets were encoded in two nibbles which couldn't have the value E or F.
The 9100B doubled the storage register count to two pages of 16, but otherwise the programming model was the same.