Bio­di­ver­sity off­sets are seen as a pol­icy mech­a­nism to bal­ance devel­op­ment and con­ser­va­tion goals. Many off­set schemes employ habi­tat restora­tion in one area to recre­ate bio­di­ver­sity value that is destroyed else­where, assum­ing that recov­ery is timely and pre­dictable. Recent research has chal­lenged these assump­tions on the grounds that restora­tion implies long time delays and a low cer­tainty of suc­cess. To inves­ti­gate these asser­tions, and to assess the strength of empir­i­cal sup­port for off­set pol­icy, we used a meta-analytic approach to ana­lyze data from 108 com­par­a­tive stud­ies of sec­ondary growth (SG) and old-growth (OG) habi­tat (a total of 1228 SG sites and 716 OG ref­er­ence sites). We extracted species check­lists and cal­cu­lated stan­dard­ized response ratios for species rich­ness, Fisher’s alpha, Soren­son sim­i­lar­ity, and Morisita-Horn sim­i­lar­ity. We mod­eled diver­sity change with habi­tat age using gen­er­al­ized lin­ear mod­els and multi-model aver­ag­ing, cor­rect­ing for a num­ber of poten­tial explana­tory vari­ables. We tested whether (1) diver­sity of pas­sively and actively restored habi­tat con­verges to OG val­ues over time, (2) active restora­tion sig­nif­i­cantly accel­er­ates this process, and (3) cur­rent off­set poli­cies are appro­pri­ate to the pre­dicted uncer­tain­ties and time lags asso­ci­ated with restora­tion. The results indi­cate that in the best case, species rich­ness con­verges to OG ref­er­ence val­ues within a cen­tury, species sim­i­lar­ity (Soren­son) takes about twice as long, and assem­blage com­po­si­tion (Morisita-Horn) up to an order of mag­ni­tude longer (hun­dreds to thou­sands of years). Active restora­tion sig­nif­i­cantly accel­er­ates the process for all indices, but the inher­ently large time lags, uncer­tainty, and risk of restora­tion fail­ure require off­set ratios that far exceed what is cur­rently applied in prac­tice. Restora­tion off­set pol­icy there­fore leads to a net loss of bio­di­ver­sity, and rep­re­sents an inap­pro­pri­ate use of the oth­er­wise valu­able tool of ecosys­tem restoration.