The specific dye gives maximum absorbance at the lambda max provided by the supplier/labelled on dye under dilute aqueous phase. However when the same dye solution at high concentration when operated for spectral scan, deviates.
Self-association of dye molecules at high concentrations could alter the absorbance spectrum. Excessive absorbance that goes beyond the dynamic range of the spectrophotometer (e.g. absorbance >3) would cause flattening of the spectrum.
Guobing Xiang probably identified the main problem: the Lambert-Beer law is valid only over a limited range of absorbance. Another problem can be that "analytical grade" in dyes means little, there may be significant contaminants. Do a thin-layer chromatography (Kieselgel, butanol/acetic acid/water) with analytical grade eosin and you'll know what I mean. Yuck!
Yes, Guobing, but remember that the literature value of the extinction coefficient may have been determined with "analytical grade" dye as well! If you want to be sure, you have to either synthesise the dye yourself, or purify the commercial material, for example by chromatography. In the case of fluorescein this is easy: just boil the fluorescein with hydrochloric acid, this removes most of the junk. With that clean material, the synthesis of eosin or erythrosin is simple.