diff --git a/include/socks/util.h b/include/socks/util.h
index ba2fd65..8205781 100644
--- a/include/socks/util.h
+++ b/include/socks/util.h
@@ -2,7 +2,9 @@
 #define SOCKS_UTIL_H_
 
 #include <stddef.h>
+#include <stdbool.h>
 
 extern void data_size_to_string(size_t value, char *out, size_t outsz);
+static inline bool power_of_2(size_t x) { return (x > 0 && (x & (x - 1)) == 0); }
 
 #endif
diff --git a/include/socks/vm.h b/include/socks/vm.h
index 741ec0b..2072b2d 100644
--- a/include/socks/vm.h
+++ b/include/socks/vm.h
@@ -137,9 +137,11 @@ typedef struct vm_cache {
 	unsigned int c_obj_size;
 	/* combined size of vm_slab_t and the freelist */
 	unsigned int c_hdr_size;
+	/* power of 2 alignment for objects returned from the cache */
+	unsigned int c_align;
 	/* offset from one object to the next in a slab.
-	   this may be different from c_obj_size as
-	   we enforce a 16-byte alignment on allocated objects */
+	   this may be different from c_obj_size depending
+	   on the alignment settings for this cache. */
 	unsigned int c_stride;
 	/* size of page used for slabs */
 	unsigned int c_page_order;
diff --git a/vm/cache.c b/vm/cache.c
index 016b49b..f4d1b54 100644
--- a/vm/cache.c
+++ b/vm/cache.c
@@ -1,7 +1,8 @@
 #include <socks/queue.h>
 #include <socks/compiler.h>
-#include <stddef.h>
 #include <socks/vm.h>
+#include <socks/util.h>
+#include <stddef.h>
 
 #define FREELIST_END ((unsigned int)-1)
 
@@ -31,13 +32,19 @@ void vm_cache_init(vm_cache_t *cache)
 		cache->c_flags |= VM_CACHE_OFFSLAB;
 	}
 
+	if (power_of_2(cache->c_obj_size)) {
+		cache->c_align = cache->c_obj_size;
+	} else {
+		cache->c_align = 8;
+	}
+
 	size_t available = vm_page_order_to_bytes(cache->c_page_order);
 	size_t space_per_item = cache->c_obj_size;
 
-	/* align to 16-byte boundary */
-	if (space_per_item & 0xF) {
-		space_per_item &= ~0xF;
-		space_per_item += 0x10;
+	/* align to specified boundary */
+	if (space_per_item & (cache->c_align - 1)) {
+		space_per_item &= ~(cache->c_align - 1);
+		space_per_item += cache->c_align;
 	}
 
 	cache->c_stride = space_per_item;
@@ -55,6 +62,15 @@ void vm_cache_init(vm_cache_t *cache)
 	cache->c_slabs_empty = QUEUE_INIT;
 
 	cache->c_hdr_size = sizeof(vm_slab_t) + (sizeof(unsigned int) * cache->c_obj_count);
+
+	/* for on-slab caches, c_hdr_size is added to the slab pointer to
+	   get the object buffer pointer. by aligning c_hdr_size to the
+	   requested alignment, we ensure that the object buffer
+	   is aligned too */
+	if (cache->c_hdr_size & (cache->c_align - 1)) {
+		cache->c_hdr_size &= ~(cache->c_align - 1);
+		cache->c_hdr_size += cache->c_align;
+	}
 }
 
 void vm_cache_destroy(vm_cache_t *cache)