Doing an analysis using something like
high_mean(yourfieldname) will restrict the finding of the anomalies on the high-side (and not anomalies for unusually low values). This will properly find the anomaly shown above. Anomalies are always scored relative to their level of unusualness, so smaller spikes are scored lower than larger spikes. You would then potentially alert only on the high scoring anomalies.
However, if you are asking for the functionality such as "find me anomalous spikes in the data, but NEVER in the case where the anomalous value is less than 1 Mbyte, then you'll need to wait for a few versions for a forthcoming feature that will allow specific rules to override/control what gets considered an anomaly.