diff --git a/lib/dns/ratelimit.ts b/lib/dns/ratelimit.ts
index 4523604..eaf803e 100644
--- a/lib/dns/ratelimit.ts
+++ b/lib/dns/ratelimit.ts
@@ -47,8 +47,11 @@ export function createRateLimiter(options: {
   maxClients?: number;
   now?: () => number;
 } = {}): RateLimiter {
-  const qps = options.qps ?? 50;
-  const burst = options.burst ?? 100;
+  // Sized for a browser rather than against one — a client here is an address,
+  // and behind one address is a laptop with tabs open, or a whole office. See
+  // the note on DEFAULT_QPS in scripts/moshpit-dns.ts.
+  const qps = options.qps ?? 200;
+  const burst = options.burst ?? 600;
   const maxClients = options.maxClients ?? 50_000;
   const now = options.now ?? Date.now;
 
diff --git a/scripts/moshpit-dns.ts b/scripts/moshpit-dns.ts
index a457746..c52283b 100644
--- a/scripts/moshpit-dns.ts
+++ b/scripts/moshpit-dns.ts
@@ -66,6 +66,29 @@ const gateway = createGatewayResolver({
   ipv6: list(env.MOSHPIT_GATEWAY_AAAA),
 });
 
+/**
+ * What one client may ask for, per second and in a burst.
+ *
+ * The limit prices reflection: a forged source address turns our answers into
+ * someone else's inbound traffic, so a client's budget is the most we will ever
+ * send a victim who never asked. It is not meant to price *browsing*.
+ *
+ * At 50/100 it priced browsing anyway. A client here is an address, not a
+ * person, and behind one address is a laptop opening tabs — or a household, or
+ * an office. A single page asks for A, AAAA and HTTPS on every hostname it
+ * touches, so a few tabs clear 100 queries in a burst without trying, and the
+ * excess is dropped rather than refused: nothing comes back, the stub waits out
+ * its timeout, and the page finishes with subresources that never resolved.
+ *
+ * These numbers are still a bound, just one drawn around a browser instead of
+ * inside it. What they let through is ~200 answers/sec to a spoofed victim,
+ * well under 100KB/s and no kind of amplifier — ANY is already refused, which
+ * is what makes the amplification factor small enough for this to be the right
+ * trade.
+ */
+const DEFAULT_QPS = 200;
+const DEFAULT_BURST = 600;
+
 // Off unless asked for: with it on, a name nobody holds under an ending the
 // legacy root does not have resolves to the gateway, which lands the visitor on
 // the pit with the name filled in. That is a funnel, and a funnel is a product
@@ -83,8 +106,8 @@ const dns = createDnsServer({
   address,
   port,
   rateLimiter: createRateLimiter({
-    qps: number(env.MOSHPIT_DNS_QPS, 50),
-    burst: number(env.MOSHPIT_DNS_BURST, 100),
+    qps: number(env.MOSHPIT_DNS_QPS, DEFAULT_QPS),
+    burst: number(env.MOSHPIT_DNS_BURST, DEFAULT_BURST),
   }),
   log: env.MOSHPIT_DNS_LOG === "queries" ? log : () => {},
 });
diff --git a/tests/dns-server.test.mjs b/tests/dns-server.test.mjs
index 9c6f098..89c1563 100644
--- a/tests/dns-server.test.mjs
+++ b/tests/dns-server.test.mjs
@@ -200,6 +200,24 @@ test("a client over its rate limit is dropped rather than answered", async () =>
   assert.equal(limiter.allow("192.0.2.2"), true, "a different client has its own budget");
 });
 
+test("a remote browser's page load fits inside one client's budget", () => {
+  // The limit prices reflection, not browsing. A client is an address, and
+  // behind one address is a laptop with tabs open: A, AAAA and HTTPS on every
+  // hostname a page touches clears 100 queries without trying. At the old
+  // 50/100 the excess was dropped in silence, which is a page that finishes
+  // with subresources that never resolved.
+  const limiter = createRateLimiter();
+  let served = 0;
+  for (let i = 0; i < 400; i++) if (limiter.allow("203.0.113.9")) served++;
+  assert.equal(served, 400, "a burst the size of a real page load is answered in full");
+
+  // Still a bound, though — an address cannot ask without limit.
+  const capped = createRateLimiter();
+  let allowed = 0;
+  for (let i = 0; i < 5000; i++) if (capped.allow("203.0.113.10")) allowed++;
+  assert.ok(allowed < 5000, "the budget is still finite");
+});
+
 test("the machine's own queries are never rate limited", () => {
   // Point a box's resolver at this and every query on it arrives from one
   // address. Under a per-client limit that made the whole machine share one