Although vegetation restoration is widely applied to reverse desertification in alpine sandy lands, its duration-dependent coordination of soil phosphorus (P) pools and biological P cycling remains unclear. We examined alpine sandy lands restored with Salix cupularis for 5, 10, 15, and 20 years on the eastern Qinghai-Tibet Plateau, using non-restored natural sandy land (CK) as the control. Systematic measurements included soil total P (TP), available P (AP), the P activation coefficient (PAC), labile, moderately labile, and non-labile P fractions, microbial biomass P (MBP), five phosphatase activities, and the relative abundances of phoC, phoD, and pqqC. Relative to CK, restoration significantly increased every measured indicator except non-labile P. After 20 years of restoration, surface-soil AP and PAC exceeded the control by 174.9% and 133.7%, respectively. Non-labile P content remained statistically unchanged, although its proportional contribution declined as TP and the more available P pools increased. Exploratory partial least-squares structural equation modeling summarized positive associations from P-cycling genes to enzymes (path coefficient = 0.980), from enzymes to active P fractions (0.893), and from active P fractions to P-availability indicators (0.800; all p < 0.01). Whereas, the direct enzyme–P availability association was not significant. This study shows that restoration duration is associated with P accumulation and coordinated microbial–enzymatic cycling that redistributes soil P toward more available pools without detectable depletion of non-labile P.