I usually label part of a whole information and bullet values with an integer approximation. However they are situations where higher precision is necessary.
A visual representation like bullet graph is usually displayed with its own scale. By doing that you already have a visual approximation, for given example, 75.8 is quickly decoded 75.
Tooltips suppose to give detailed information for hovered area. Using a label and a tooltip to display the same value with different levels of precision is a redundancy you do not need. In case you do so, you decode from graph 75, from label 76, from tooltip 75.8 and maybe from raw data 75.7584 (accessed by clipboard or so). If values need to be displayed, they have to have the precision required by data logic, purpose of design and media display. A tooltip can provide more detailed information for the hovered item, but is no point in using it only for two levels of precision of the same information. If data is rendered on static media (image, PDF, paper), where tooltips are not available, then greater precision might be necessary to display the values.
If for instance you have several bullet graphs arranged within a table sorted by values, sharing the same scale, you might need to display a higher precision in order to explain the ranking of some items for which the values are visually very similar.
I doubt that it has any relevance that certain isolated value is at 75.8% from target (for me 75% looks even more memorable than 76), but extra precision might have relevance for 4 consecutive values like 74.7, 74.9, 75.3, 75.5.