Standard deviations will work better if you have periods where the range of values is wider than other period - an example might be more varying response times during busy times, as opposed to quieter times.
Comparing against a Standard Deviation will reduce false alarms in those scenarios compared with a comparison against baseline percentages.
If the range of values is constant over time, always distributed over the same range, then a Baseline Percentage would work just as well.
To demonstrate this, I would recommend finding an example in your Metric Browser of a series where the baseline changes by a small amount from hour to hour, but the range varies widely - this should also have larger shaded Standard Deviation values. I've attached an example - we don't want the spike at 7:50am to generate an alarm as we know performance can be a little erratic first thing in the morning. It is well over the baseline though.