[Dis]claimer: I’m the author/maintainer of xlrd.
The documentation references to formula text are about “name” formulas; read the section “Named references, constants, formulas, and macros” near the start of the docs. These formulas are associated sheet-wide or book-wide to a name; they are not associated with individual cells. Examples: PI maps to =22/7, SALES maps to =Mktng!$A$2:$Z$99. The name-formula decompiler was written to support inspection of the simpler and/or commonly found usages of defined names.
Formulas in general are of several kinds: cell, shared, and array (all associated with a cell, directly or indirectly), name, data validation, and conditional formatting.
Decompiling general formulas from bytecode to text is a “work-in-progress”, slowly. Note that supposing it were available, you would then need to parse the text formula to extract the cell references. Parsing Excel formulas correctly is not an easy job; as with HTML, using regexes looks easy but doesn’t work. It would be better to extract the references directly from the formula bytecode.
Also note that cell-based formulas can refer to names, and name formulas can refer both to cells and to other names. So it would be necessary to extract both cell and name references from both cell-based and name formulas. It may be useful to you to have info on shared formulas available; otherwise having parsed the following:
B2 =A2
B3 =A3+B2
B4 =A4+B3
B5 =A5+B4
...
B60 =A60+B59
you would need to deduce the similarity between the B3:B60 formulas yourself.
In any case, none of the above is likely to be available any time soon — xlrd priorities lie elsewhere.